skillforge
Health Gecti
- License — License: NOASSERTION
- Description — Repository has a description
- Active repo — Last push 0 days ago
- Community trust — 17 GitHub stars
Code Gecti
- Code scan — Scanned 4 files during light audit, no dangerous patterns found
Permissions Gecti
- Permissions — No dangerous permissions requested
Bu listing icin henuz AI raporu yok.
Quality-gated skill authoring for Claude Code. Build, judge, and ship skills that earn their tokens.
Skill Forge ⚔️
Quality-gated skill authoring for Claude Code. Build, judge, and ship skills that earn their tokens.
Skill Forge is four skills that run in sequence (recap, create or update, judge, apply changes with human-in-the-loop), each pass tightening the prompt until it earns a passing grade. Enter the cycle at any point and repeat until the result is sharp.
skill-forge-judge and skill-forge-hitl also work standalone: judge any prompt, step through any numbered list.
Skills
skill-forge-recap: reads a skill's body independently of its description. Reports what the skill actually does, flags undeclared behaviors, and verdicts drift as aligned / minor / significant. Offers context-sensitive next steps: fix frontmatter via HITL, hand off to skill-forge-update, or run skill-forge-judge.
skill-forge-create: builds a new skill from scratch: discovery recap to identify domain and failure modes, pattern selection, spec-fetching draft phase with knowledge-delta discipline enforced, then judge + HITL quality gate before install.
skill-forge-update: structured update workflow for existing skills: recap, drift check, change elicitation with consistency checks, applies via HITL, judges the result, then saves and activates.
skill-forge-judge: evaluates any LLM-consumed prompt against a dimensional rubric (knowledge delta, anti-patterns, usability, spec compliance). Outputs a letter grade, per-dimension scores, and a numbered improvements list.
skill-forge-hitl: Human-in-the-Loop: steps through any numbered list one item at a time. Shows a status board upfront, applies each change, prompts approve/revise/skip, and commits after each approval.
Install
The skills ship as a single bundled plugin — create and update call judge and hitl as part of the quality gate, and recap feeds into update. One install gives you the full forge cycle:
/plugin marketplace add WrathZA/skillforge
/plugin install skillforge
Principles
See principles.md. Short version: knowledge delta not tutorials, single-keypress menus, no cross-skill file deps, every NEVER needs WHY + INSTEAD, and skills are never finished. Judge repeatedly, not just when something breaks.
Local Development
Clone the repo and run symlink-global-skills.sh once to link skills globally on this machine:
bash symlink-global-skills.sh
Symlinks each skill directory into ~/.claude/skills/, making them available to any Claude Code session regardless of working directory. Edits to the repo are immediately reflected. Safe to re-run: already-linked entries are skipped.
Credits
- Matt Pocock (aihero.dev):
write-a-skill, the original skill this collection builds on - softaworks/agent-toolkit:
skill-judge, the original eval skillskill-forge-judgeis based on
Yorumlar (0)
Yorum birakmak icin giris yap.
Yorum birakSonuc bulunamadi